abseil-cpp

308 Commits

9 Branches

22 Tags

14 MiB

Tag: Branch: Tree: 540e2537b9

Commit Graph

Author	SHA1	Message	Date
Abseil Team	389ec3f906	Export of internal Abseil changes. -- 636137f6f0de910691a3950387fefacfa4909fb8 by Abseil Team <absl-team@google.com>: Add move semantics to absl::container_internal::CompressedTuple PiperOrigin-RevId: 225394165 -- 43da91e4f95a196b2e6b76f1c2f4158817b0ebb0 by Greg Falcon <gfalcon@google.com>: Add a constructor to allow for global absl::Mutex instances. This adds a new constexpr constructor to absl::Mutex, invoked with the absl::kConstInit tag value, which is intended to be used to construct Mutex instances with static storage duration. What's tricky about is absl::Mutex (like std::mutex) is not a trivially destructible class, so by the letter of the law, accessing a global Mutex instance after it is destroyed results in undefined behavior. Despite this, we take care in the destructor to not invalidate the memory layout of the Mutex. Using a kConstInit-constructed global Mutex after it is destroyed happens to work on the toolchains we use. Google relies heavily on this behavior internally. Code sanitizers that detect undefined behavior are able to notice use-after-free of globals, and might complain about this pattern. PiperOrigin-RevId: 225389447 -- 7b553a54bc6460cc7008b028552e66799475ca64 by Abseil Team <absl-team@google.com>: Internal change. PiperOrigin-RevId: 225373389 -- fd0c722d217b3b509102274765ccb1a0b596cf46 by Abseil Team <absl-team@google.com>: Update absl/time/CMakeLists.txt to use new functions i.e. absl_cc_(library\|test) PiperOrigin-RevId: 225246853 -- 9f8f3ba3b67a6d1ac4ecdc529c8b8eb0f02576d9 by Abseil Team <absl-team@google.com>: Update absl/synchronisation/CMakeLists.txt to use new functions i.e. absl_cc_(library\|test) PiperOrigin-RevId: 225237980 -- a3fdd67dad2e596f804f5e100c8d3a74d8064faa by Abseil Team <absl-team@google.com>: Internal cleanup PiperOrigin-RevId: 225226813 -- 48fab23fb8cdca45e95da14fce0de56614d09c25 by Jon Cohen <cohenjon@google.com>: Use a shim #define for wchar_t in msvc in int128. On ancient versions of msvc and with some compatibility flags on wchar_t is a typedef for unsigned short, whereas on standards-conforming versions wchar_t is a typedef for __wchar_t. The first situation causes int128 to not compile as you can't define both `operator wchar_t()` and `operator unsigned short()` because they are the same type. This CL introduces a wrapper #define in order to abstract over the different typedefs for wchar_t. We do a define instead of a typedef so that we can #undef at the end and not leak the symbol, since we need it in a header. https://docs.microsoft.com/en-us/previous-versions/dh8che7s(v=vs.140) has more detail about the underlying problem. PiperOrigin-RevId: 225223756 GitOrigin-RevId: 636137f6f0de910691a3950387fefacfa4909fb8 Change-Id: Iad94e52e9484c5acec115a2f09ef2d5ec22c2074	6 years ago

Author

SHA1

Message

Date

Abseil Team

389ec3f906

Export of internal Abseil changes.

--
636137f6f0de910691a3950387fefacfa4909fb8 by Abseil Team <absl-team@google.com>:

Add move semantics to absl::container_internal::CompressedTuple

PiperOrigin-RevId: 225394165

--
43da91e4f95a196b2e6b76f1c2f4158817b0ebb0 by Greg Falcon <gfalcon@google.com>:

Add a constructor to allow for global absl::Mutex instances.

This adds a new constexpr constructor to absl::Mutex, invoked with the absl::kConstInit tag value, which is intended to be used to construct Mutex instances with static storage duration.

What's tricky about is absl::Mutex (like std::mutex) is not a trivially destructible class, so by the letter of the law, accessing a global Mutex instance after it is destroyed results in undefined behavior. Despite this, we take care in the destructor to not invalidate the memory layout of the Mutex. Using a kConstInit-constructed global Mutex after it is destroyed happens to work on the toolchains we use. Google relies heavily on this behavior internally.

Code sanitizers that detect undefined behavior are able to notice use-after-free of globals, and might complain about this pattern.

PiperOrigin-RevId: 225389447

--
7b553a54bc6460cc7008b028552e66799475ca64 by Abseil Team <absl-team@google.com>:

Internal change.

PiperOrigin-RevId: 225373389

--
fd0c722d217b3b509102274765ccb1a0b596cf46 by Abseil Team <absl-team@google.com>:

Update absl/time/CMakeLists.txt to use new functions
i.e. absl_cc_(library|test)

PiperOrigin-RevId: 225246853

--
9f8f3ba3b67a6d1ac4ecdc529c8b8eb0f02576d9 by Abseil Team <absl-team@google.com>:

Update absl/synchronisation/CMakeLists.txt to use new functions
i.e. absl_cc_(library|test)

PiperOrigin-RevId: 225237980

--
a3fdd67dad2e596f804f5e100c8d3a74d8064faa by Abseil Team <absl-team@google.com>:

Internal cleanup

PiperOrigin-RevId: 225226813

--
48fab23fb8cdca45e95da14fce0de56614d09c25 by Jon Cohen <cohenjon@google.com>:

Use a shim #define for wchar_t in msvc in int128.

On ancient versions of msvc and with some compatibility flags on wchar_t is a typedef for unsigned short, whereas on standards-conforming versions wchar_t is a typedef for __wchar_t. The first situation causes int128 to not compile as you can't define both `operator wchar_t()` and `operator unsigned short()` because they are the same type.

This CL introduces a wrapper #define in order to abstract over the different typedefs for wchar_t. We do a define instead of a typedef so that we can #undef at the end and not leak the symbol, since we need it in a header.

https://docs.microsoft.com/en-us/previous-versions/dh8che7s(v=vs.140) has more detail about the underlying problem.

PiperOrigin-RevId: 225223756
GitOrigin-RevId: 636137f6f0de910691a3950387fefacfa4909fb8
Change-Id: Iad94e52e9484c5acec115a2f09ef2d5ec22c2074

6 years ago

1 Commits (540e2537b92cd4abfae6ceddfe24304345461f32)